Structural insights into Charcot–Marie–Tooth disease‐linked mutations in human GDAP1

Charcot–Marie–Tooth disease (CMT) is the most common inherited peripheral polyneuropathy in humans, and its different subtypes are linked to mutations in dozens of different genes. Mutations in ganglioside‐induced differentiation‐associated protein 1 (GDAP1) cause two types of CMT, demyelinating CMT4A and axonal CMT2K. The GDAP1‐linked CMT genotypes are mainly missense point mutations. Despite clinical profiling and in vivo studies on the mutations, the etiology of GDAP1‐linked CMT is poorly understood. Here, we describe the biochemical and structural properties of the Finnish founding CMT2K mutation H123R and CMT2K‐linked R120W, both of which are autosomal dominant mutations. The disease variant proteins retain close to normal structure and solution behavior, but both present a significant decrease in thermal stability. Using GDAP1 variant crystal structures, we identify a side‐chain interaction network between helices ⍺3, ⍺6, and ⍺7, which is affected by CMT mutations, as well as a hinge in the long helix ⍺6, which is linked to structural flexibility. Structural analysis of GDAP1 indicates that CMT may arise from disruption of specific intra‐ and intermolecular interaction networks, leading to alterations in GDAP1 structure and stability, and, eventually, insufficient motor and sensory neuron function.

Inherited polyneuropathies are a genetically and clinically diverse group of neurodegenerative diseases affecting motor and sensory neurons in the peripheral nervous system (PNS) [1,2]. Mutations in dozens of genes expressed in the PNS cause Charcot-Marie-Tooth disease (CMT). Based on clinical findings, CMT can be classified into three forms: demyelinating, axonal, and intermediate [3,4]. The progress of CMT is linked to the hereditary pattern, whereby the autosomal recessive form has an earlier onset and more severe symptoms than the autosomal dominant form [5][6][7]. Understanding the molecular function of the proteins involved in the etiology of neuropathies is vital in efforts toward treatment and diagnosis.
Ganglioside-induced differentiation-associated protein 1 (GDAP1) is an integral mitochondrial outer membrane (MOM) protein, and the GDAP1 gene is one of the most abundant targets for missense mutations linked to CMT [8][9][10]. Both autosomal dominant and recessive modes of inheritance are found, resulting in either autosomal recessive or dominant demyelinating CMT4, autosomal dominant axonal CMT2, or intermediate CMTRIA types of CMT, with varying phenotype severity [11]. The mutations R120W and H123R, which we focus on in this study, are both autosomal dominant mutations causing the CMT2K subtype. Both phenotypes show typical slow development after onset, and main symptoms include loss of sensation in limb extremities and muscle weakness. The clinical profiling of the phenotypes has been described earlier in Spain and Finland [12][13][14]. GDAP1 is ubiquitously expressed in tissues, but most of the expression confines to neuronal tissues [9,15]. In the cell, GDAP1 localizes as a tailanchored MOM protein [16]. Structurally, GDAP1 resembles glutathione S-transferases (GST), and it contains unique flexible loops [17,18]. The most accurate structural data thus far cover the dimeric core GST-like domain of human GDAP1, including the GDAP1specific insertion [18]. The transmembrane helix and the GST-like domain are linked by a hydrophobic domain and possibly a flexible linker loop.
Glutathione S-transferase superfamily members function in prokaryotic and eukaryotic metabolism through the utilization of reduced glutathione to catalyze a range of chemically diverse reactions. GSTs often contribute to mechanisms of neurodegenerative diseases [19,20]. In comparison with many other enzyme superfamilies, GSTs are unique in that sequence conservation appears to be driven by fold stability instead of catalytic features, as reflected in the broad spectrum of GST substrates [21,22]. While the function of GDAP1 is not fully understood at the molecular level, it has been linked to multiple mitochondrial events in neurons [23,24], redox regulation, and signal transduction [25,26].
In the Finnish population, the autosomal dominant founder mutation H123R accounts for as much as 20-30% of the local CMT cases [12,27]. We carried out structural analysis of two selected autosomal dominant GDAP1 mutants, H123R and R120W, using X-ray crystallography and complementary biophysical and computational techniques. In addition, we used three cell culture models, rat dorsal root ganglion (rDRG) neurons, human embryonic kidney 293 (HEK293T) cells, and human skin fibroblasts, to observe the oligomeric state of GDAP1 and the effects of the disease mutations therein.

Cloning
The GDAP1D295-358 and GDAP1D303-358 constructs used to produce soluble recombinant human GDAP1 in E. coli have been described [18]. Point mutations were generated in GDAP1D303-358 by a site-directed mutagenesis protocol with Pfu polymerase [28]. An N-terminal His 6 -affinity tag and a tobacco etch virus (TEV) protease digestion site were included in each construct. The full-length GDAP1 coding sequence was subcloned into Gateway Ò (Invitrogen) vectors pEN-TTmcs and pSLIK-HYGRO [29]. Point mutations were introduced as above. In addition to the GDAP1 gene, the tetracycline-responsive promoter element (tight-TRE) was added within the cloning site [30], and a single Nterminal FLAG-tag was introduced into each construct [31]. All constructs were verified by DNA sequencing.
For GDAPD295-358, the Ni 2+ -NTA purification protocol was identical, but 40 mM HEPES, 400 mM NaCl, 20 mM imidazole, and pH 7.5 were used as lysis and Ni-NTA as washing buffer, and 32 mM HEPES, 320 mM NaCl, 500 mM imidazole, and pH 7.5 were used to elute bound proteins. EDTA-free protease inhibitor cocktail (Roche) was included during cell freezing and lysis. TEV protease treatment was performed in dialysis against 40 mM HEPES, 400 mM NaCl, and pH 7.5 at +4°C overnight, followed by a second Ni 2+ -NTA affinity step. SEC was performed using a Superdex 200 16/60 HiLoad column (Cytiva) with 20 mM HEPES, 300 mM NaCl, 1% (v/v) glycerol, 0.5 mM TCEP, and pH 7.5 as mobile phase. SEC peak fractions were analyzed with SDS-PAGE, and Coomassie-stained bands were used for protein identification using a Bruker UltrafleXtreme matrix-assisted laser desorption/time-of-flight mass spectrometer (MALDI-TOF-MS). Tryptic peptides extracted from the gel were identified by a search in NBCI and Swiss-Prot databases using BIO-TOOLS2.2 (Bruker, Billerica, MA, USA).
Crystallization, data collection, and structure determination Mutant GDAP1D303-358 crystals were obtained using the sitting-drop vapor-diffusion method at +4°C. Proteins were mixed with mother liquor on crystallization plates using a Mosquito LCP nanodispenser (SPT Labtech, Melbourn, UK). The protein concentration was 10-30 mgÁmL À1 in 75 nL, and 150 nL of reservoir solution was added. H123R crystals were obtained in 0.15 M DLmalic acid and 20% (w/v) PEG3350. R120W crystals were obtained in 0.1 M HEPES (pH 7.3) and 10% (w/v) PEG6000. Crystals were briefly soaked in a mixture containing 10% PEG200, 10% PEG400, and 30% glycerol for cryoprotection, before flash freezing in liquid N 2 .
A novel crystal form of wild-type GDAP1D295-358 was obtained at +8°C in 200 mM NH 4 formate, 25% (w/v) PEG3350 in a drop containing 150 nL of 8.64 mgÁmL À1 protein, and 150 nL of reservoir solution. Cryoprotection was performed by adding 3 µL of cryoprotectant solution [75% (v/v) reservoir solution mixed with 25% (v/v) PEG200] directly into the crystallization drop, followed by crystal mounting and flash freezing with liquid N 2 .
Diffraction data collection at 100 K was conducted at the PETRA III synchrotron source (DESY, Hamburg, Germany) on the P11 beamline [33,34] and the EMBL/-DESY P13 beamline. Diffraction data were processed and scaled using XDS [35]. Crystal structures of wild-type GDAP1D303-358 [18] were used as search models in molecular replacement. Molecular replacement, model refinement, and structure validation were done using Phenix [36,37] and CCP4 [38]. The models were refined using PHENIX.REFINE [39] and rebuilt using COOT [40]. The structures were validated using MOLPROBITY [41]. The data processing and structure refinement statistics are shown in Table 1.

Modeling, simulation, and bioinformatics
A model for full-length GDAP1 was obtained from Alpha-Fold2 [42] and used for further analyses as such. In addition, crystal structure-based models were prepared and analyzed. Missing loops of the wtGDAP1 crystal structure were built with YASARA [43], and the structure was minimized. The model was further used as a starting point for small-angle X-ray scattering (SAXS) data fitting (see below), as well as molecular dynamics (MD) simulations.
MD simulations were run on a GDAP1 monomer, with all loops in place. The simulations were run using GRO-MACS [44], with input file preparation on CHARMM-GUI [45]. The force field used was CHARMM36m [46], with a cubic box and a 10-A extension around the protein. Solvation was done with the TIP3P water model in 0.15 M NaCl. Temperature (NVT) equilibration to 300 K and pressure (NPT) equilibration, via isotropic pressure coupling, were carried out using the Berendsen thermostat. Structural properties of GDAP1 were analyzed with bioinformatics tools, including NAPS [47] for centrality analyses and DynaMine [48] for prediction of flexibility. Stability effects of missense mutations were predicted with CUPSAT [49] and MAESTRO [50,51]. Hydrophobic clusters were identified with PROTEINTOOLS [52].

Small-angle X-ray scattering
The structure and oligomeric state of the GDAP1 R120W and H123R mutants were analyzed with SEC-SAXS on the SWING beamline [53] (SOLEIL synchrotron, Saint Aubin, France). Samples were dialyzed against fresh SEC buffer and centrifuged at > 20 000 g for 10 min at +4°C to remove aggregates. 50 µL of each protein sample at 8.5-10 mgÁmL À1 was injected onto a BioSEC3-300 column (Agilent, Santa Clara, CA, USA) at a 0.2 mLÁmin À1 flow rate. SAXS data were collected at +15°C, over a q-range of 0.003-0. 5 A À1 (q = 4p sin(h)/k, where 2h is the scattering angle).
Further processing and modeling were done using ATSAS 3.0 [54]. Scattering curves were analyzed and particle dimensions determined using PRIMUS [55] and GNOM [56], respectively. Chain-like ab initio models were generated using GASBOR [57]. Different GDAP1 dimer models were fitted in a complementary approach against the experimental SAXS data using CRYSOL [58]. SUPCOMB was used to superimpose SAXS models and crystal structures [59].

Circular dichroism spectroscopy
Synchrotron radiation circular dichroism (SRCD) spectra were collected from 0.5 mgÁmL À1 samples on the AU-CD beamline at the ASTRID2 synchrotron source (ISA, Aarhus, Denmark). The samples were prepared in a buffer containing 10 mM HEPES, pH 7.5, and 100 mM NaF. The samples were equilibrated to room temperature and applied into 0.1-mm pathlength closed circular quartz cuvettes (Suprasil, Hellma Analytics, M€ ullheim, Germany). SRCD spectra were recorded from 170 nm to 280 nm at +25°C. Three repeat scans per measurement were recorded and averaged. The CD spectrum baselines were processed and converted to molar ellipticity using CDTOOLX [60].

Thermal stability
Thermal unfolding of GDAP1 variants in SEC buffer was studied by nanoDSF using a Prometheus NT.48 instrument (NanoTemper, Munich, Germany). Tryptophan fluorescence was excited at 280 nm, and emission was recorded at 330 and 350 nm. The samples were heated from +20 to +90°C with a heating rate of 1°CÁmin À1 , and changes in the fluorescence ratio (F350/F330) were monitored to determine apparent melting points. The data were analyzed using Origin (Origi-nLab Corporation, Northampton, MA, USA).

Cell culture and western blotting
Human skin fibroblast cultures were established from skin biopsies of a healthy donor and a patient with GDAP1 H123R mutation [12,61]. Written consent for the use of patient material was obtained, and the study was approved by the Coordinating Ethics Committee of the Helsinki and Uusimaa Hospital District.
The purification of rDRG sensory neurons, the generation of lentiviral particles, and their use to overexpress the GDAP1 constructs were done as described [62] for MORC2. All animal work was performed in accordance with the Swedish regulations and approved by the regional ethics review committee Stockholm, Stockholms norra djurf€ ors€ oksetiska n€ amnd (N79/15). rDRGs were matured into sensory neurons, and GDAP1 expression was induced by doxycycline-initiated tight-TRE promoter expression using the lentivirus system [30].
The protein fractions were isolated from the rDRG and fibroblast cells and membranes using 40 mM HEPES, pH 7.0, 400 mM NaCl, and 1% n-dodecyl-b-D-maltopyranoside, and the supernatant was clarified by centrifugation at 235 000 g, +4°C. The proteins were separated with 12% SDS-PAGE under nonreducing conditions.
HEK293T-D10 cells were used to serve as endogenous control and to test the redox sensitivity of the mammalianderived GDAP1 samples. Proteins were isolated from total cell lysate and mitochondrial fraction. The mitochondria were isolated from the cells using linear 15-50% (w/v) sucrose gradient centrifugation. The protein was treated with similar lysis conditions as above, and SDS-PAGE was performed with and without 192 mM b-mercaptoethanol.

Structural effects of CMT mutations H123R and R120W on GDAP1
The CMT-linked missense mutations in GDAP1 are clustered within the vicinity of the hydrophobic clusters of the N-terminal GST-like domain (GSTL-N), the C-terminal GST-like domain (GSTL-C), and the dimer interface. The affected side chains are often polar or charged and orient toward the solvent (Fig. 1A). They are also close to the hydrophobic clusters of GDAP1 (Fig. 1B). For example, the a6 helix, Lys188-Glu229, has 20 charged residues along the helix. The clustered mutations could change the sidechain interaction networks between helices a3, a6, and a7, which further might affect GDAP1 folding and stability. Here, we focused on two CMT-linked GDAP1 mutations on helix a3 pointing toward a6, R120W, and H123R (Fig. 1A).
We previously determined the crystal structure of wild-type GDAP1Δ303-358, which corresponds to the GST-like core domain of GDAP1 in dimeric form, including the GDAP1-specific insertion [18]. Here, we expressed and purified the variants R120W and H123R, compared them with wild-type GDAP1 (wtGDAP1) crystal and solution structures, and studied their folding and thermal stability. The crystal of the R120W mutant variant had a new crystal form, while H123R had the same space group as the wtGDAP1 structure, displaying a homodimer in the asymmetric unit. In the H123R structure, the dimer in the crystal is covalently linked via a disulfide bond at Cys88, like the wtGDAP1 protein [18]. The disulfide bridge via Cys88 also exists in the R120W structure, but the dimer is formed via crystallographic symmetry.
Both Arg120 and His123 are on the a3 helix, partially solvent-accessible (Fig. 1C,D). In both mutant structures, as in wtGDAP1, the most flexible regions are in loops between b3-b4 at positions Leu71-Ala77, and a5-a6 at positions Arg159-Ile186. The b3-b4 loop is more structurally ordered in the mutants than in wtGDAP1. The a5-a6 region corresponds to the GDAP1-specific insertion in the GST superfamily [63]. In both mutant structures, the flexible loop between helices a6-a7 is similar to wtGDAP1; the Ca backbone is visible, but side chains have poor density.
In the crystal state, the mutations do not cause major structural changes ( Fig. 1D-F). However, intramolecular interactions are altered. In the H123R structure (Fig. 1E), the His123-Tyr124 p-orbital interaction is disturbed in the mutant, while the interaction with the side chain of Gln218 is preserved. The salt bridge network around Glu222 and Arg226 is conserved and now includes Arg123. In chain B, the electron density for Arg123 is weak, indicating the flexibility of the mutant residue.
Arg120 in wtGDAP1 forms a H-bond with the backbone carbonyl of Cys240, and it is part of a salt bridge network involving Glu222, Arg226, and Glu229 (Fig. 1F). Intriguingly, Arg120 has close contact with Arg226 in wtGDAP1, whereby the two Arg p systems stack, and the surrounding Glu222 and Glu229 neutralize charges via salt bridges. Trp120, as a bulky side chain, causes steric hindrance in the R120W mutant (Fig. 1F), and the a3 helix, carrying Trp120, moves outwards by~1 A, and the contact with the neighboring a6 is weakened. The backbone interaction with Cys240 is lost in the mutant, and the salt bridge network centered at Arg226 is disturbed as is the contact between His123 and Gln218, which could be linked to the loss of protein stability. Arg226 turns into the generated pocked, making a weak hydrogen bond to the backbone carbonyl of Cys240 at the expense of its normal interactions (Fig. 1G).

The mutant proteins show unaltered conformation but lowered stability
To compare the solution and crystal structures, SAXS analysis was performed on the H123R and R120W mutants (Fig. 2). SEC-SAXS was employed to achieve better separation between monomer and dimer fractions. Previously, we showed this equilibrium to be concentration-dependent; high concentration favors the dimeric form [18]. The SEC-SAXS profiles show that the samples are monodisperse with the same radius of gyration, R g , across the peak (Fig. 2A). The molecular weight across the main peak showed that in both mutants, the peak contained a dimeric form similar to wtGDAP1; accordingly, the scattering curves for all variants were essentially identical (Fig. 2B).
Further analysis revealed that the R g values matched the ones for wtGDAP1 (Table 2), indicating similar solution conformation and oligomeric state. Both mutants possess a similar globular fold with essentially the same level of flexibility as wtGDAP1, as demonstrated by the dimensionless Kratky plots (Fig. 2C). Distance distribution functions revealed that the particle dimensions in solution are nearly identical between the mutants and wtGDAP1 (Fig. 2D). Hence, at the resolution of a SAXS experiment, neither mutation caused large-scale conformational changes.
The dimer-monomer equilibrium in GDAP1 is dynamic, and the dimeric form is favored at high concentrations [18]. At the concentrations and conditions used here, GDAP1 exists as a dimer, as indicated by the SAXS data. Chain-like ab initio models confirmed the observation that both mutants are nearly indistinguishable from dimeric wtGDAP1 (Fig. 2E,F). Thus, there is no indication of effects on the oligomeric status by the two mutations.
The SAXS data indicate a dimeric form for both mutants ( Table 2). The ambiguity between the two mutants was estimated with AMBIMETER [64], and H123R and R120W have both similar levels of globularity and stability. These observations are in line with the fairly minor conformational differences in the crystal state, whereby R120W-as a more drastic replacement-led to a small movement of the a3 helix and loss of hydrogen bonding interactions. On the contrary, the only fully monomeric GDAP1 mutant we studied earlier, Y29E/C88A, is more globular than any of the dimeric forms [18].
To further compare the molecules in solution and in the crystal, the crystal structure coordinates were fitted against the SAXS scattering curve (Fig. 2E). Based on the analysis, both mutants adopt the dimer form, having very similar folds in solution as the wtGDAP1. After building in the loops, the dimer structure fits the data better than the crystal structure (see below); hence, the dimer observed in the crystal represents the solution structure, with the addition of the GDAP1specific flexible insertion.
Since the mutants presented similar solubility and folding as wtGDAP1, we tested whether the mutations caused changes to GDAP1 stability or secondary structure content. To test for quantitative differences in secondary structures, we measured SRCD spectra (Fig. 3A). The SRCD spectra of the mutants overlay well with the wtGDAP1 CD spectrum, showing that the mutants, on average, have a similar secondary structure composition in solution as wtGDAP1. The CD peak at 208 nm is weaker for H123R, which may indicate minor differences in intramolecular interactions between ⍺-helices, as seen in the crystal structure. Once the GDAP1 dimer forms via the disulfide bond, the structure becomes very stable, and it is challenging to dissociate the dimer [18]. Due to the high helical content in GDAP1 and the similarity of the CD spectra, we can confirm that the mutations do not affect the overall folding characteristics of GDAP1. Small differences in spectral shape may be caused by both local stacking of amino acid side chains and interactions between secondary structure elements [65,66].
To determine thermal stability, we studied the R120W and H123R variants using the Trp fluorescence emission peak ratio at wavelengths 350/330 nm in nanoDSF. The wtGDAP1 protein is more stable than the mutants (Fig. 3B). The apparent T m value for wtGDAP1,~62°C, was >5°C higher than for both mutants, suggesting that the effect of the mutations may be linked to an overall destabilization of the fold. Considering the location of the mutations, a region of GDAP1 is revealed, which is important for protein stability.
To obtain further insight into the effects of the mutations on GDAP1, we used a variety of bioinformatics tools. Analyses of centrality (Fig. 3C) indicated that the core region of GDAP1, close to both Arg120 and His123, with helix ⍺7 the most central element, is likely to be important for folding and stability. Many other CMT mutations cluster into this area [18], affecting a number of residues in an interaction network (Fig. 1C). Effects of point mutations on protein stability against temperature or chemicals were predicted using CUPSAT [49]. Both R120W and H123R are predicted to be destabilizing in both respects, in line with the thermal stability data above. DynaMine analysis (Fig. 3D) of protein flexibility based on sequence data further showed that the part of helix ⍺6 before residue 200 has context-dependent rigidity, indicating that the unique ⍺5-⍺6 loop in GDAP1 is structurally dynamic.
To complement the above analysis, we predicted the effects of all possible mutations in human GDAP1 on the protein stability, using the crystal structure of wild-type GDAP1 [18] as a starting point. Using two alternative prediction algorithms [49][50][51], we observed an overall destabilizing effect predicted for the CMT-linked GDAP1 mutations. Only~10% of all CMT mutations were predicted to be stabilizing by both programs, and individually, both methods predicted 70-80% of the known mutations to decrease GDAP1 stability. No clear difference was observed between dominant and recessive mutations in this respect, but the numeric values of ΔΔG indicate a slightly stronger overall destabilizing effect for the dominant mutations. The results of the predictions are summarized in Table 3, and the raw data can be found in Data S1.

Flexibility of the a6 helix
A crystal structure was solved for wtGDAP1D295-358, and it presented a novel crystal form with four monomers in the asymmetric unit. One complete dimer was present in the asymmetric unit, in addition to two half-dimers, which both homodimerize through crystallographic symmetry. As the resolution of the structure was rather low, structural details were not analyzed. However, in all four independent protomers, helix a6 breaks in the middle around Asp200, and a helix at residues 189-198 is present in electron density (Fig. 4A). Thus, the long a6 helix can adopt different conformations even in the crystal state. The wtGDAP1 crystal structure published earlier (Fig. 1A) had an asymmetric dimer, with one short and one long a6 helix [18]. These observations suggest that the GDAP1-specific insertion is also flexible in solution.  Table 3. Predicted effects of CMT missense mutations on GDAP1 stability, using the CUPSAT and MAESTRO servers. The structure used for the predictions was the A chain of wild-type GDAP1, PDB entry 7ALM. Only residues present in the crystal structure were considered. Note that for CUPSAT, a negative sign for ΔΔG corresponds to destabilization, while a positive ΔΔG is destabilizing in MAESTRO. The new wtGDAP1 dimer structure was analyzed with respect to the SAXS data, together with a dimer built based on the AlphaFold2 model monomer. The AlphaFold2 model has the helix a6 divided into two and collapsed into a similar, but even more compact, conformation as seen in the new wtGDAP1 crystal (Fig. 4B). However, it is evident from data fitting that the AlphaFold2 model is too compact, while the extended conformation of helix a6 is too elongated.

All mutations
An excellent fit to the SAXS data was obtained using the dimer from the new wtGDAP1 crystal structure with built-in missing loops (Fig. 4C-E).
To further analyze the dynamics of GDAP1, the model based on the wtGDAP1 crystal structure [18], with all loops added, was subjected to MD simulations (Fig. 5). Throughout the simulation, the GDAP1-specific insertion is the most dynamic segment of the protein, but the long helix remains extended. On the contrary, simulation

GDAP1 mutant localization and oligomeric state in cells
To explore the two CMT-linked mutations, R120W and H123R, at the cellular level, we used three different cell culture models, in which either endogenously expressed GDAP1 variants or inducible systems to overexpress GDAP1 variants were utilized.
The localization of R120W and H123R in neurons was compared to that of wtGDAP1 using rDRG primary cell cultures. After protein induction, the neurites were immunostained and imaged using a confocal microscope (Fig. 6). FLAG-tagged wtGDAP1 locates both in the cell body and in the axons of the sensory neurons. The localization is not cytoplasmic or on the plasma membrane and, in accordance with previous work, most likely mitochondrial. No apparent difference in the immunostaining intensity or cellular localization of R120W or H123R was observed compared with wtGDAP1. None of the mutations induced distinguishable morphological changes in sensory neurons, and no cell toxic effects were observed (Fig. 6).
Dimeric GDAP1 has been detectable from mammalian sources [67]. To further test whether full-length GDAP1 preferably forms disulfide-linked dimers in cells, we performed western blot analysis for protein extracts under nonreducing conditions (Fig. 7). We analyzed rDRG sensory neurons overexpressing fulllength GDAP1 variants (Fig. 7A,B) and HEK293T-D10 cells that endogenously express GDAP1 (Fig. 7D, E) to explore the oligomeric state of wtGDAP1, R120W, and H123R. In all cases, nonreducing SDS-PAGE and western blotting showed only a monomeric form. We then used human fibroblast cultures established from skin biopsies of a CMT2K patient carrying the H123R GDAP1 allele [12,61]. Normal GDAP1 genotype fibroblasts were used as control. Western blot analysis revealed that in the fibroblasts, full-length GDAP1 exists as monomers in both CMT2K patientderived cells and healthy control samples (Fig. 7C). For comparison, purified recombinant GDAP1 core domain (lacking the transmembrane domain) and its mutant variants were predominantly dimeric under nonreducing SDS-PAGE (Fig. 7F).
In all cell lysates studied here, full-length wtGDAP1, as well as both mutant variants thereof, was detected as a monomer after electrophoresis (Fig. 7). Hence, any dimer present in the cells in these experiments, in the presence of the transmembrane domain, is not disulfide-linked. Of note, the mutation C88A did not make the recombinant GDAP1 GST-like domain fully monomeric in our earlier study [18], indicating that the homodimer of the folded core domain can form without the disulfide bridge in vitro. The only GDAP1 variant we have observed to be fully monomeric is the double mutant Y29E/C88A [18], disturbing both the hydrophobic interface and the disulfide bridge. To conclude, the main form of full-length GDAP1 in cells, at least in the absence of inducing factors, is not a disulfide-linked dimer.

Discussion
Single-amino-acid substitutions can alter protein physicochemical properties, affecting protein stability and function. From a clinical perspective, inherited neuropathies are generally well characterized at the level of the symptomatic spectrum and disease progression. A large variety of CMT mutations are known and characterized clinically [68]. However, in many cases, the molecular basis of these disorders cannot be adequately explained. The difficulty of understanding the mechanism is due to both the vast number of the involved genes and their heterogeneous inheritance patterns and phenotypes, as well as limited knowledge about molecular structure and function. GDAP1 is one of the genes associated with peripheral neuropathies caused by missense mutations. We performed structural analyses for two human CMT2K-linked GDAP1 mutations on helix ⍺3, which revealed that apart from the mutated residue and its immediate surroundings, the overall fold does not change. However, both mutations introduce changes in intramolecular networks and differences in molecular properties, most notably in thermal stability. The structural analysis of pathogenic CMT-linked GDAP1 variants shows that the mutations are close to the GDAP1 hydrophobic cluster and mediate interactions between key helices of the structure.

CMT mutations in GDAP1 cluster into hot spots in 3D space
Currently, there are at least 103 GDAP1 mutations linked to CMT, out of which 68 are reported missense mutations [69]. The functional effects of several GDAP1 mutations have been studied in cells using neurons and Schwann cells or yeast models [7,10,70,71]. As the actual function of GDAP1 at the molecular level, including possible enzymatic activity, remains unclear, additional insights can be obtained using structural biology techniques.
We chose to investigate the Finnish H123R founder mutation, as well as R120W, due to its wellestablished clinical and molecular characterization and its location in the vicinity of His123, on helix ⍺3. Both mutations have been linked to the autosomal dominant form CMT2K. Both R120W and H123R are common mutations in European patients [7,14,72], and H123R was identified as a major founder mutation in the Finnish population [12,13].
The GDAP1 crystal structure allows predicting the molecular basis for many of the known mutations in the Human Gene Mutation Database (http://www. hgmd.cf.ac.uk/ac) and the Inherited Neuropathy Variant Browser database (https://neuropathybrowser. zuchnerlab.net). A CMT-related mutation cluster of GDAP1 (Fig. 1) localizes more on helices a3 and a6 and less on helices a7, a8, and their connecting loops [18]. There are 68 known missense GDAP1 mutations involving 39 residues. The main cluster contains 27 residues that form a network of interactions, including salt bridges, hydrogen bonds, and van der Waals contacts. These interactions provide extensive contacts between helices ⍺3, ⍺6, and ⍺7 (Fig. 1C). Centrality analyses of the GDAP1 structure highlight this, indicating helix ⍺7 as the most central segment of the GDAP1 structure. Notably, many mutations linked to CMT2K map very close to each other in 3D space, suggesting an intramolecular network that may get disturbed upon disease mutations, altering GDAP1 structure or function.
Mutating residues His123 and Arg120 does not break the GDAP1 fold, but rather may affect the intramolecular residue interaction network and protein stability. The thermal stability of the mutant proteins decreased compared with wtGDAP1, suggesting that especially the interaction between helices a3 and a6 may be important. Predictions of ΔΔG using computational methods mainly agree with the experiment, suggesting that both mutations are destabilizing; overall, CMT mutations are predicted to destabilize GDAP1 (Table 3). Interestingly, the wedge between helices a5 and a6 contains a pattern of double tyrosines, double glutamates, double leucines, and double lysines, which seem to pull the GSTL-C core together. The mutations in many cases are introduced into the neighboring positions of these double pairs, like in the case of H123R.
We showed that hexadecanoic acid bound into a groove close to the CMT mutation cluster [18]. The R120W and H123R mutation sites are close to the a6 helix, and the main hydrophobic cluster centered around a7. As an interesting hot spot, Arg120 is the site for four different CMT mutations. Mutations in such clusters and hot spots might affect residue interaction networks and thus decrease protein stability.

The network of interactions is sensitive toward CMT mutations
Considering the interactions between His123 and Arg120, as well as the networks between helices ⍺3, ⍺6, and ⍺7, it becomes evident that several involved residues are targets of CMT mutations. An example is Glu222, which is sandwiched between three Arg side chains (Arg120, Arg225, and Arg226) and Tyr124, and has van der Waals contacts to Leu239. Another example is Ala247; the conservative mutation to valine is linked to disease [73]. Ala247 on helix ⍺7 is located in a tight space right below His123 and Arg120 and part of the same interaction network. The apparently mild CMT mutation A247V increases the size of the side chain and affects the local interactions. Cys240 from the ⍺6-⍺7 loop also lies right below His123 and Arg120, and its mutation to Tyr in CMT [70,74] will interfere with the local interaction network at this residue cluster.
Taken together, although the CMT mutations in GDAP1 initially appear to be scattered throughout the sequence, in the 3D structure, they are often involved in close, complex interaction networks, and these networks are sensitive toward changes in many different participating residues. This observation, together with our bioinformatics analyses, suggests the loss of protein stability upon mutations in such networks and clusters and may hint at an overall mechanism of GDAP1-linked CMT. Similarly, we earlier observed a decrease in protein stability for all known CMT-linked mutations in the myelin protein P2 [75,76]. It should, however, be noted that the exact linkage of the observed decrease in protein stability caused by CMT mutations in vitro and in silico to actual alterations in GDAP1 protein function in vivo remains currently unclear.
A key feature of GDAP1-not present in the construct used for structural studies here-is a single transmembrane helix, which anchors into the MOM. The residues linked to the most severe phenotypes of CMT locate in the close vicinity or inside the transmembrane helix [77], hence not participating in the interaction networks within the folded GDAP1 core domain, while the vast majority of the mutation sites cluster into the folded GST-like domain (Fig. 8). Our data show that the R120W and H123R mutations had only minor effects on the crystal structure, while decreasing the GDAP1 fold stability. Such effects might be altered once the protein is fully constituted into a membranea question that could be answered through structural studies on full-length GDAP1 constructs including the transmembrane domain. The severe CMT mutation G237D results in a mistargeted protein due to the inclusion of a negative charge into the transmembrane helix [77]. Interestingly, Gly-rich sequences, also called glycine zippers, are often involved in transmembrane helix dimerization [78], and the Gly motif in the transmembrane domain could be involved in GDAP1 dimerization. However, we did not find any disulfide-linked dimeric GDAP1 in cell lysates, containing the endogenous or overexpressed full-length GDAP1. Although the functional effects of a missing transmembrane domain or some common missense point mutations have been described [16,79,80], more comprehensive modeling [81] coupled with complete structural data from full-length GDAP1 would better explain the mild phenotype mutations. After optimization of large-scale production of full-length recombinant human GDAP1 and its mutant variants, future work on GDAP1 structure will include the above aspects.
The unique ⍺6 helix of GDAP1 Helix ⍺6 is a dominant and unique feature of the GDAP1 structure, being part of the GDAP1-specific insertion, together with its preceding loop, which is not visible in electron density. We used a combination of crystal structures and computational models to get further insights into the GDAP1 helix ⍺6 and its dynamics. Our observation of the a6 conformational flexibility may point out mechanistically important functions, which could be linked to the effects of CMT disease mutations on or near the helix.
In the original wtGDAP1 structure [18], we saw a breakdown of noncrystallographic symmetry, as the ⍺6 helix was of different lengths between the two chains in the asymmetric unit. The shorter version of the helix started around residue 200, which is the hinge region in our new wtGDAP1 crystal structure, in which the helix continues in another direction in all four independent protomers in the asymmetric unit. A break of the helix at the same position is also predicted by AlphaFold2. However, the conformation of monomeric GDAP1 in the model is incompatible with the exact model of dimerization we observe in the crystal state, leading to steric clashes. Sequence-based flexibility analysis points out the region around residues 190-200 as potentially flexible. The conformational dynamics of the GDAP1-specific insertion, via a hinge around residue 200, could play a role in its physiological functions and its interactions with other molecules, such as the cytoskeleton, in vivo.
GDAP1 is a dimer in solution but not disulfidelinked in cells GDAP1 has a unique dimer interface compared with canonical GSTs [18] and dynamic oligomerization properties [18,67,82]. Our results show that GDAP1 dimerization mediated via a disulfide bond can also be observed in the CMT mutant proteins in vitro. In the cellular environment, such a disulfide bond could be formed via a folding catalyst or through changes in the redox environment. The latter has been linked to GDAP1 function [83].
An interesting possibility for the dimeric function would be GDAP1 activation by a folding catalyst, affecting interactions with a partner protein, suggesting that the GDAP1 function could be linked to its oligomeric state. Many binding targets have been proposed for GDAP1, such as tubulin and other cytoskeletal components [23,71]. The GDAP1 disease mutations could, in addition to affecting protein folding and stability, modulate protein-protein interactions. However, the details of molecular interactions formed by GDAP1 remain unknown, and further studies should be focused on studying GDAP1-cytoskeleton interactions and their links to GDAP1 oligomeric state.

Conclusions
We have presented a structural analysis of two CMTlinked mutations in GDAP1, R120W, and H123R. The effects of these mutations on protein structure were small, and the mutations may affect dynamic properties, stability, and conformational changes of GDAP1 and/or its interactions with additional binding partners. The cluster of CMT-related mutations in the 3D structure of GDAP1 highlights a tightly interwound network of amino acid side-chain interactions that are likely essential for the normal function and structure of GDAP1. Such mutation clustering establishes the need for accurate structural studies of proteins targeted by disease mutations, and it can be expected that most of the mutations in such a cluster similarly affect protein stability or functional interaction networks. A major goal for the future will be the structure solution of full-length GDAP1 protein, including the transmembrane domain, in addition to deciphering details of its physiological function and molecular interactions.
acknowledged. Furthermore, we wish to thank Dr. Roman Chrast for project planning and organization.